`scope_for_offset` does not make sense

We have scope_for_offset function, which finds expression's scope based on offset. It is used primaraly for completion, where offset is the postion of the cursor.

Currently, ExprScopes is implmented by maintaining a tree of scopes (= sets of names) and a mapping from expressions to scopes. At the moment, scope_for_offset works by, roughly, finding the nearest expression to the offset and using it's scope. The problem with this approach is that it doesn't work. here's fun example

fn main() {
  let x;
  let y;
  /* caret here */
  let z;
}

Here, we obviously want to see x and y in scope, but not z. The problem is, there are no expressions in the snippet at all, so the "find nearest expression" plan fails.

Proposed solution is to store scope origin along with each scope, and, instead of finding closest expression, look for "closest scope", where "closes scope" is actually a bunch of hand-written heuristics for various edge cases, plus the base rule about enclosing expression.

Start at the scope_for_offset function and the ExprScopes implementation described in the issue. Reproduce the completion case with the cursor before let z, then determine the scope-origin and closest-scope behavior needed so x and y are visible while z is not.